Reinforcement learning post-training drives reasoning and agentic capabilities in modern AI systems, yet a growing body of work shows that it is most effective when used to fine-tune an already capable base model. We question whether existing pipelines yield models that are most suitable for reinforcement learning. Building on prior work highlighting the role of coverage and pass@K as predictors of post-RL performance, we design a simple modification to supervised fine-tuning, TailSFT, which filters out already fit sequences during training, thereby focusing learning on under-modeled regions, or the tail, of the data distribution. We justify and validate the design choices in TailSFT, particularly the specific filtering criteria, through a combination of controlled experiments and theoretical analysis. On OLMo-3 7B, TailSFT often improves pass@16 performance on math and coding evaluations, with gains up to 17% absolute, while incurring minimal computational overhead. These higher-coverage checkpoints consistently translate to up to 4% absolute pass@1 gains in subsequent GRPO runs, demonstrating that TailSFT checkpoints serve as better initializations for RL. We further introduce a lightweight diagnostic for identifying settings where TailSFT is most likely to help. More broadly, our results motivate a principled, stage-aware approach to model development, in which intermediate checkpoints are judged by how effectively they support subsequent training.
Many machine-learning systems set a threshold at a quantile of a calibration set: conformal predictors that promise 90% coverage by drawing their cutoff at the calibration set's 90th percentile, abstention gates that decline to answer when a model's score falls below the calibration set's tenth percentile, safety filters that block any output scoring above the 99th percentile of a reference set. All of them promise that the threshold will hold at the stated rate on new data. The promise assumes the calibration examples are independent, and in modern pipelines they usually are not: they share a prompt, a document, a reasoning trace. Survey statistics has known how to discount correlated data since 1965, by counting how many independent observations a sample is worth, but only for averages. We show that a threshold needs a different count. The count depends on how often clustered scores land on the same side of the threshold, and that changes with where the threshold is set. How similar the scores are as numbers does not enter. We prove a closed-form law for the resulting effective sample size and for the spread of the coverage a deployed system actually sees. Three consequences follow. The correction now used in the conformal literature is the wrong quantity, and can miss in either direction. A dataset has no single effective sample size. It has one for each level the threshold is set at. And the damage is invisible in coverage averaged over many runs, and fully felt by whoever deploys once. On a released calibration set of 25,028 examples, we measure the reliability of about 1,300.
While autonomous coding agents have significantly advanced automated test generation, they remain fundamentally limited by lazy generation, a phenomenon where agents prematurely terminate tasks and systematically avoid complex programmatic logic, resulting in inadequate code coverage. Currently, mitigating this premature termination requires continuous human-in-the-loop supervision. This heavy reliance on human intuition creates a bottleneck that negates the efficiency gains of automated generation. We propose SCATE, a framework for adaptive, automated supervision of coding agents that replaces human intervention during test generation. By formulating supervision as a contextual bandit problem, SCATE learns to select the most promising testing actions based on the current coverage and class testability metrics, maximizing coverage gains while minimizing wasted generation effort. Our empirical evaluation demonstrates that SCATE integrates seamlessly with different coding agents. When applied to GEMINI-CLI, it achieves 32.3% higher line coverage and 30.9% higher branch coverage than the agent-only baseline. A comparison with CLAUDE CODE confirms the framework dynamically adapts its policy to optimize each agent's unique strengths. SCATE also consistently outperforms state-of-the-art non-agentic approaches across all metrics.
Modern processor verification struggles to reach deep architectural states due to the inefficiencies of traditional mutation-based fuzzing. We propose HiFuzz, a novel hierarchical reinforcement learning framework that replaces mutation with a structured, two-layer generation process: a Program Agent for global layout and a Basic Block Agent for precise instruction filling. To overcome reward sparsity, HiFuzz integrates an adaptive coverage reward mechanism and a semantic-aware basic block encoder providing intrinsic feedback. Extensive evaluations on three real-world RISC-V cores demonstrate that HiFuzz significantly outperforms state-of-the-art fuzzers in coverage and bug detection.
Learning systems based on IF-THEN rule representations readily offer interpretability, making them a crucial focus in contemporary AI research. A key objective for such rule sets is to achieve both high discriminative power and interpretability. While existing state-of-the-art algorithms implicitly prioritize predictive accuracy, they often fall short on one or more quality metrics that ensure interpretability, such as coverage and parsimony of rule sets. Motivated by this, this paper propose the development of CDPR, which aims to create highly accurate and interpretable rule sets for classification problems. To the best of our knowledge, this represents the first attempt to establish such an approach. In this study, we introduce two algorithms rooted in submodular maximization, which not only provide provable guarantees on coverage but also yield rule sets that are both discriminative and parsimonious. We empirically demonstrate that rule sets learned through our approaches achieve higher accuracy and interpretability and has more than a 2.5-fold improvement in average coverage rates when compared to the next best algorithm.
Deliberative polling promises to improve collective decision-making by exposing shareholders to a broad range of arguments before they vote. Yet ensuring that every voter encounters a representative sample of the reason space, the coverage problem, remains an open challenge, particularly at scale and in adversarial or strategically motivated electorates. This paper introduces a way of evaluating solutions using the LLM-based Agentic Bipolar Argumentation Simulator, grounded in a framework which formalises a poll as a six-tuple <Jend, Jopp, Ratt, Renh, VA, VR> of endorsing and opposing justifications, attack and enhance relations, and shareholder- and relation-weights. ABAS simulates N autonomous shareholder agents, each assigned a latent opinion according to desired distributions in [-1, 1], who sequentially vote, choose or author justifications, and optionally submit argumentation-graph links. The simulator implements recommendations that rank existing justifications by their observable endorsement mass. It evaluates the mechanism's success by coverage, namely the fraction of the corpus reason-tag set represented in the K recommendations presented to each shareholder, as a solution to the NP-hard Subsuming Justification Problem. Reported experiments characterise how creativity rate (pown), recommendation size (K), argumentation density (plinks), and population size (N) affect coverage and corpus diversity. In an authenticated electorate where Sybil attacks are impossible and only the relation graph is gameable, we stress-test the scoring with coordinated strategic voting attacks: a tag-flood attack collapses coverage, while author-count relation weighting through a reversed-PageRank rule resists the flood markedly better than uniform weights.
Reference-free faithfulness metrics verify each atomic claim a model makes against ground truth, and are increasingly used to evaluate grounded generation. We show they share a blind spot: they measure only precision -- are the stated claims supported? -- and therefore reward abstention, since a model can score near-perfect faithfulness by saying almost nothing. We make this measurable using Formula 1 telemetry, a domain where strategic ground truth is derived deterministically and, crucially, completely: for each decision we know the full set of facts that mattered. This completeness -- absent in open-domain faithfulness benchmarks -- lets us measure recall (coverage of the relevant facts) exactly, alongside precision. On a multilingual (EN/ES/PT) benchmark of 7,253 decision instances spanning 150 races, the most precise frontier model covers under half of the relevant facts and ranks last by F1, so requiring coverage reorders the systems; the same effect reappears in a second complete-oracle domain (NOAA weather forecasts). A prompt ablation shows the low coverage is not an under-prompting artifact: explicitly asking models to be thorough does not close the gap. We pair faithfulness with coverage into a single score, validate the metric (controlled perturbation; agreement across a model-free regex extractor and a cross-family LLM extractor, system-level Spearman 1.0), and give a verifier-guided generation method that improves precision and recall without references. We release the benchmark, structured annotations, metric, baselines, and an interactive demo.